Thyroid cancer is the most common type of cancer of the endocrine system, and its incidence has increased almost 3-fold over the past decades.  Traditional clinical characteristics, such as TNM stage, can be used to predict the severity related to TC, but it is difficult to accurately estimate the risk of recurrence.  Thus, it is imperative to establish powerful tools that can be effectively applied to aid in the diagnosis, prognosis, and treatment of patients with TC.
Accumulating evidence shows that since lncRNAs have high tissue and cancer specificity, they might play an active role in cancer initiation, development and progression. An increasing number of studies have shown that lncRNAs promote tumor cell proliferation, invasion, metastasis, and angiogenesis and can serve as an excellent tool to modulate therapeutic decisions in cancer. Increasing evidence suggests that lncRNAs are involved in TC tumorigenesis and progression as important regulatory factors, thus, lncRNAs have attracted much attention as potential targets in the diagnosis and treatment evaluation of TC. For example, Liu et al. found that MALAT1 may have an oncogenic function in PTC and may thus be a potential diagnostic marker for PTC . In our research, a few of the DEirlncRNAs in the model, such as SMIM25,  LINC00900,  AC010980.2,  HAGLROS, and LINC01977 , have already been revealed to play roles in various cancers, especially TC, while others were found to be related to TC for the first time. Nevertheless, it is necessary to validate whether this immune-related lncRNA model could be a helpful predictive indicator in TC. Many researchers are currently focused on constructing signatures with both coding genes and noncoding RNAs, which can assess the survival status of patients with malignant carcinoma[31, 32]. Unlike most traditional risk models, our newly constructed signature involved two-lncRNA pairwise comparisons and relative ranking on the basis of gene expression entirely from the same TC patient. Although from different sequencing platforms, our prognostic model does not require gene expression data normalization. Previous studies have supported the effectiveness of this method[33, 34].
In the current study, we established an immune-related lncRNA model and evaluated its prognostic value as well as its correlation with immune cell infiltration, ICI-related genes and TKIs in TC. First, we performed differential coexpression analysis to identify DEirlncRNAs based on data from TCGA. LncRNA pairs were systematically identified through pairwise comparisons in the same sample without the need for data normalization. In addition, univariate analysis with Lasso regression analysis was performed on the pairs to validate the most suitable variables. Fourteen significant DEirlncRNA pairs with maximum prognostic values were determined with multiple repeats and random stimulation. Next, we used these pairs to develop the predictive risk score model. Then, we calculated not only the 3-, 5-, and 10-year AUC values of the ROC curve but also identified the optimal cutoff point of the 1-year ROC curve to separate TC patients into high- and low-risk groups. Furthermore, Kaplan-Meier curves, time-dependent ROC curves, and Cox proportional hazards regression analysis showed that the model has independent predictive value for TC prognosis. Finally, we evaluated the relationship between this novel model and tumor-infiltrating immune cells, ICI-related molecules and small-molecule inhibitor validity.
Immune infiltrates in the tumor microenvironment (TME) play a vital role in tumor development and progression and affect the clinical outcomes of cancer patients.  Dysfunction of the immune status in the TME contributes to the development and progression of cancer, and this is the basis of many immunotherapy studies. Moreover, tumor immunotherapy is now considered to have an important role in the elimination of cancer cells and sheds light on the mechanisms of cancer–immune evasion, contributing to tumor outgrowth. Recent studies have suggested that lncRNAs play a central role in innate and adaptative cancer immunity regulation.  Immune-related lncRNA pairs as signatures are better at predicting prognosis than single lncRNAs. Therefore, it is necessary to explore more immune-related lncRNAs in tumors for future clinical practice. In this research, we carried out pairwise comparisons of a given set of immune-related lncRNAs and expression values. Thus, our prognostic signature could help address batch effects between different platforms and overcome the reprocessing and normalization of data.
Immune cell infiltration reflects the TME and reportedly impacts the outcome of TC progression. It is evident that immune-related lncRNAs are correlated with the development of TC. To explore the relationship between the prognostic model and immune-infiltrating cells, we applied seven commonly accepted methods, including TIMER[37, 38], CIBERSORT, XCELL, QUANTISEQ , MCP-counter,  EPIC  and CIBERSORT-ABS . By integrating analyses, we found that the levels of Tregs, myeloid dendritic cells, monocytes, cancer-associated fibroblasts and B cell plasma in the high-risk group were higher than those in the low-risk group, while the levels of neutrophils, M1 macrophages, CD8+, CD4 + T cells, and B cells were significantly negatively correlated with the risk of signature. These results were consistent with the findings of some previous experimental studies, which aimed to determine the correlation between each cell type and the aggressiveness of TC[45–47]. For example, it has been reported that neutrophils play an antitumor role and can be beneficial to the prognosis of TC, which is consistent with the findings of our analysis. Our results also revealed that the abundance of Tregs was more associated with the high-risk group, which is similar to the findings of previous literature. Those studies found that the levels of Tregs in PTC were higher than those in multinodular goiter patients, and Tregs were consistently present in extraglandular invasion and lymph node metastasis[49, 50]. Monocytes have been observed to promote the occurrence and development of tumors, and their high density is closely related to thyroid tumor invasion and reduced survival, which was also confirmed by our study. These observations can be further explored for a holistic understanding of the nuances of TC microenvironment immune cell infiltration.
Tyrosine kinase inhibitors (TKIs) are an innovative personalized strategy that aim at pro-oncogenic kinases, including EGFR, MET, PDGFR, VEGFR-1, VEGFR-2, RAF, FGFR and RET. Our signature showed that the risk score was related to some of these inhibitors, such as gefitinib, sunitinib, and tipifarnib, indicating that this new model might be a new method in the assessment of efficacy to systemic therapy based on a genetic understanding in TC. Moreover, ICB immunotherapy is viewed as a promising cancer therapeutic modality for malignant tumors. The identification of PD-L1 as an immunostat blockade has led to the development of a number of cancer immunotherapies. For RAI refractory PTC patients, recent evidence has shown that overexpression of PD-L1 together with lymphocyte infiltration into the tumor TME are significantly associated with the effectiveness of ICB[52, 53]. In this study, we found that a high-risk score was negatively correlated with ICB-related genes such as PD-1, PD-L1, LAG3, CTLA-4, PD-L2 and CD74, which are commonly expressed in human cancer. In addition, some clinical trials have applied PD-1 and PD-L1 inhibitors in combination with TKIs, RAIs or chemotherapy for managing and defeating deadly TC. Our signature may provide new insight to predict which patients are more suitable for these treatments, either alone or in combination.
To the best of our knowledge, a prognostic model based on irlncRNA pairs in TC has not been reported to date. Our predictive model is based on a 0-or-1 matrix and could be applied in an individualized manner while eliminating batch bias. In addition, our signature first combined DElncRNA pairs with ICB and TKI efficacy for analysis. Various additional methods were used to support the prognostic value and feasibility of this new model.
However, this lncRNA-based prognostic signature had several limitations. First, the establishment and validation of the model was based only on the TCGA database, which might lead to selection bias. To verify the predictive values of the risk assessment model, a larger dataset and external datasets of TC should be analyzed. Second, this was a retrospectively designed study, and a prospective cohort needs to be established in future work for further verification. In addition, the calculation formulas of this prognostic signature may be too complex for clinical application.